23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66

Incomplete lineage sorting (ILS) and introgression increase genealogical discordance across the genome, which complicates phylogenetic inference. In such cases, identifying orthologs that result in gene trees with low estimation error is crucial because phylogenomic methods rely on accurate gene histories. We sequenced whole genomes for the tinamous (Aves: Tinamidae) to dissect the sources of gene and species-tree discordance and reconstruct their interrelationships. We compared results based on four ortholog sets: (1) coding genes (BUSCOs), (2) ultraconserved elements (UCEs) with short flanking regions, (3) UCEs with intermediate flanks, and (4) UCEs with long flanks. We hypothesized that orthologs with more phylogenetically informative sites would result in more accurate species trees because the resulting gene trees contain lower error. Consistent with our hypothesis, we found that long UCEs had the most informative sites and lowest rates of error. However, despite having many informative sites, BUSCO gene trees contained high error compared to long UCEs. Unlike UCEs, BUSCO gene sequences showed a positive association between the proportion of parsimony informative sites and gene tree error. Thus, BUSCO and UCE datasets have different underlying properties of molecular evolution, and these differences should be considered when selecting loci for phylogenomic analysis. Still, species trees from different datasets were mostly congruent. Only one clade, with a history of ILS and introgression, exhibited substantial species-tree discordance across the different data sets. Overall, we present the most complete phylogeny for tinamous to date, identify a new species, and provide a case study for species-level phylogenomic analysis using whole-genomes.

Although the growth of high-throughput sequencing approaches over the past decade has greatly improved our understanding of evolutionary relationships, reconstructing the tree of life remains an enduring challenge. Analyses that utilize alternative datasets, methodological frameworks, or substitution models to answer the same phylogenetic questions often yield conflicting results, which is surprising given that phylogenomic datasets often contain hundreds of thousands or even millions of phylogenetically informative characters (Dunn et al. 2008; Jarvis et al. 2014; Prum et al. 2015; Reddy et al. 2017; Simion et al. 2017; Franz et al. 2019; Schultz et al. 2023) . In many cases, alternative phylogenomic topologies resulting from different methods or data are equally well-supported. Thus, two key questions for molecular biologists are, (1) why do phylogenomic datasets yield discordant topologies? and (2) how should one choose among conflicting well-supported topologies?

One way in which discordant phylogenies can result from the same or similar datasets is from the use of different methods for reconstructing species trees. One frequently employed method is to concatenate all alignments into a single supermatrix and treat the resulting phylogeny as a genome-wide average (henceforth, concatenation). Another common approach is to estimate a gene tree for each molecular marker independently and summarize their histories to estimate the species tree based on expectations from a multi-species coalescent (MSC) model (Zhang and Mirarab 2022) . Unlike the concatenation approach, MSC methods are designed to account for expected variation in gene histories due to incomplete lineage sorting (ILS). ILS causes some gene trees to differ from the species tree due to retained ancestral 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 variation during speciation, resulting in alleles coalescing prior to speciation events in ways that result in incongruence between gene and species trees (Maddison 1997) .

Each of these two methods, concatenation and MSC, has theoretical advantages, but can result in erroneous species trees under certain conditions. For example, there is an expectation that concatenation may be the best method when ILS is weak but is likely to be statistically inconsistent when ILS is strong (Mendes and Hahn 2018; Bryant and Hahn 2020) . This is because when ILS is strong, gene genealogies are expected to differ more from the species tree (i.e., there is increased heterogeneity) compared to situations when ILS is weak, thus reducing the probability that concatenation will find the true tree. ILS is more likely to occur when successive divergence times are short and effective population sizes are large, such as during rapid or adaptive radiations (Maddison 1997; Mclean et al. 2019; Lescroart et al. 2023; Tan et al. 2023) . Thus, discordance among concatenation and MSC methods may be in part driven by the presence of rapid diversification that leads to ILS and high gene tree heterogeneity that biases concatenation results.

When applying MSC methods, identifying orthologous markers that lead to accurate, unbiased gene tree estimation is crucial to properly infer phylogeny (Meiklejohn et al. 2016; Springer and Gatesy 2016; Tea et al. 2021) . MSC methods that utilize gene trees as input (gene tree summarization methods) may fail if gene tree heterogeneity is driven by estimation error rather than coalescence events. In other words, the MSC model is intended to incorporate gene tree heterogeneity due to ILS, but erroneous gene trees resulting from low quality, biased, or inconsistent sequence data will increase species tree estimation error (Xi et al. 2015) . Thus, dissecting biological and artifactual sources of gene tree discordance to more accurately infer species trees is a major challenge in the age of phylogenomics, where more data does not automatically translate to improved inferences (Meiklejohn et al. 2016; Blom et al. 2017; Smith et al. 2023) . Because MSC methods assume that all gene tree discordance is driven by ILS, eliminating erroneous gene trees becomes important (Roch and Warnow 2015; Springer and Gatesy 2016). This point has led to a variety of methods for filtering genomic datasets for phylogenetic efficacy (Doyle et al. 2015; Kuang et al. 2018; B.T. Smith et al. 2018; S.A. Smith et al. 2018; Zhao et al. 2023) but has generally lacked consensus on the types of data that contain low error to begin with.

Given the complexities associated with phylogenetic inference, identifying datasets of gene orthologs containing minimal bias is a principal goal of systematics research and has implications across diverse fields in molecular biology. One widely employed data type for phylogenomics includes coding sequences obtained via anchored hybrid enrichment, transcriptomics, or low coverage whole-genome sequencing (Lemmon et al. 2012; Prum et al. 2015; Allen et al. 2017; Waterhouse et al. 2017; Johnson 2019; Zhang et al. 2019; Burbrink et al. 2020; Boyd et al. 2022; Van Damme et al. 2022) . Another data type includes ultraconserved elements (UCEs), which target conserved genomic regions to identify orthology, but also yield sequences of their more variable flanking regions for phylogenetic reconstruction across a range of timescales (Faircloth et al. 2012; McCormack et al. 2012; Faircloth et al. 2013; Musher and Cracraft 2018; Gueuning et al. 2020; Catanach et al. 2021; Ostrow et al. 2023) . Both data types are widely applied in phylogenomics but can sometimes result in different phylogenetic topologies. For instance, coding and non-coding markers have resulted in conflicting topologies for the backbone of Neoaves (McCormack et al. 2013; Jarvis et al. 2014; Prum et al. 2015) , and 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 some authors have suggested that this discordance can be attributed to data biases associated with coding genes (e.g., elevated GC content and model misspecification) that also bias downstream species tree inferences (Reddy et al. 2017). Nevertheless, there have been few studies comparing the efficacy of these marker types to one another. Thus, the ability of coding versus primarily non-coding markers to resolve phylogenomic trees remains an open question.

One method for evaluating ortholog quality and gene tree error is through quantification of alignment information content, such as the number of parsimony informative sites (henceforth, PIS), or by comparing levels of gene tree heterogeneity among data types (Fong and Fujita 2011; Harris et al. 2014; Burbrink et al. 2020; Leite et al. 2021; Smith et al. 2023) . For example, alignments with few PIS often result in erroneous or poorly resolved gene trees because of a lack of phylogenetic signal in the data (Xi et al. 2015; Meiklejohn et al. 2016) . In such cases, we expect datasets with fewer PIS to result in increased gene tree heterogeneity that is driven by gene tree estimation error. For example, alignments of coding genes, or UCEs that include long flanking regions, should contain many PIS, and therefore should result in lower rates of gene tree error than UCEs with short flanking regions that contain relatively few variable sites. Some studies have found that loci with more PIS result in more <clock-like= gene trees, suggesting that more informative loci result in gene trees with branch lengths that represent real biological signal rather than data-driven artifacts (Musher et al. 2019) . Other studies have shown a negative correlation between the number of PIS and Robinson-Foulds distance (a measure of phylogenetic dissimilarity; henceforth, RF distance) between gene and species trees (Burbrink et al. 2020) . This relationship implies that more informative alignments result in reduced gene tree estimation error. Alternatively, one might also expect that phylogenetic markers that are too variable can result in erroneous gene trees, if phylogenetic signal becomes lost due to sequence saturation (Felsenstein 1978; Brinkmann et al. 2005) . Thus, documenting the relationship between alignment information content and RF distance, as well as the variation in overall RF distances among different datasets, should inform future phylogenomic study design and illuminate data-type efficacy.

Whole-genome sequencing offers a way to test the robustness of a phylogenetic topology given multiple types of phylogenetic markers, including both coding and non-coding sequences, from the same underlying data. For example, whole genomes are often used to harvest large numbers of single-copy orthologous protein-coding markers called BUSCO (Benchmarking Universal Single Copy Orthologs) genes (Waterhouse et al. 2017; Alaei Kakhki et al. 2023) . Given their conservation of function, BUSCO genes can theoretically tackle both relatively deep and shallow phylogenetic problems (Timilsena et al. 2022; Van Damme et al. 2022; Alaei Kakhki et al. 2023) . Similarly, there are multiple pipelines that can easily be used to obtain UCEs and other conserved elements for phylogenomics from whole-genome datasets (Faircloth 2016; Edwards et al. 2017) . Thus, sequencing whole genomes enables a robust test of the efficacy of different data types for phylogenomics.

In this study, we apply whole-genome sequencing to reconstruct a species-level phylogeny of the tinamous (Aves: Tinamidae). There has been limited work on tinamou phylogenomics to date. Most past phylogenetic studies utilized either morphological data with relatively few genetic markers (Bertelli et al. 2002; Bertelli and Porzecanski 2004; Valqui 2008; Bertelli 2017; Almeida et al. 2022) , or phylogenomic data with large-scale molecular sampling but minimal taxonomic sampling (Cloutier et al. 2019) . Thus, the species-level phylogenetic 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 relationships of the tinamous remain uncertain. Here, we build on past studies by sampling thousands of orthologous markers across 45 out of 46 described species and a total of 65 named taxa (monotypic species plus subspecies). Specifically, we compare levels of PIS and gene tree estimation error in two types of orthologous markers: complete single-copy proteincoding genes (BUSCO9s) and UCE9s. In doing so, our objectives are to (1) examine the effect of alignment information content on gene tree estimation error, (2) use this information to identify high quality (low error) orthologs, (3) identify the drivers of phylogenetic discordance among data types and methods, and (4) use these inferences to robustly reconstruct the evolutionary history of the tinamous.

Here we used whole-genome sequencing to explore the biological and artifactual sources of phylogenomic conflict and reconstruct the species-level history of a relatively old Neotropical avian group of broad interest in evolutionary biology (Bertelli and Porzecanski 2004; Prum et al. 2015; Altimiras et al. 2017; Sackton et al. 2019; Li et al. 2023) , the tinamous. Although we found that alignment information content and gene tree error varied considerably within and among datasets, our results based on different analytical approaches (concatenated and MSC methods) and datasets (UCEs and BUSCOs) were largely congruent. The topology of only a single clade (Clade A) varied among methods and datasets (Figures 1, 2, and 3), indicating that species tree estimation is, in general, robust to gene tree estimation error when biological sources of gene tree heterogeneity are limited. Nevertheless, the concatenated tree shows that Clade A contained multiple successive short internodal branch lengths (Figure 1), which suggests ILS may be a key driver of gene tree heterogeneity in this clade, and thus an important factor driving phylogenetic conflict. Indeed, quartet analysis revealed multiple internal branches with relative quartet frequencies approaching the ⅓ threshold predicted by high ILS (Figures 7A and S7). Cases such as this, where ILS is strong, necessitate the use of datasets without significant gene tree estimation error to accurately infer the species tree because concatenation is expected to fail when ILS is strong (Roch and Warnow 2015; Xi et al. 2015; Springer and Gatesy 2016) . To complicate matters, we found evidence of historical introgression in this clade that may have further elevated levels of gene tree discordance (Figure 7B). For this reason, we suggest the MSC tree based on the UCE1000Flank dataset, which had the lowest rates of gene tree estimation error (lowest median and variance in RF distances between gene and species trees) is likely the most reliable (Xi et al. 2015) . Nevertheless, in addition to ILS, rapid divergence events can be difficult to reconstruct because the short timeframe also means there is limited phylogenetic signal for those divergence events within individual gene trees ( Leaché et al. 2015 ; Springer and Gatesy 2016; Mclean et al. 2019). Thus, given the lack of agreement across datasets and methods, we would argue that more analysis is needed to confirm the relationships of this clade that diversified rapidly with introgression between non-sister taxa.

The relationship between alignment information content and gene tree error 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373

We examined the effects of data type (coding versus primarily non-coding datasets) and information content (number and proportion of PIS) on gene tree estimation error and found that coding sequences and genes with relatively few PIS contained high rates of gene tree estimation error. Although UCEs conformed to the expected negative association between PIS and gene tree estimation error as measured using RF distance (Figure 6), BUSCOs did not. Instead, we recovered a positive association between the proportion of PIS per alignment and RF distance, possibly because coding genes with more variable sites are evolving faster and thus are more prone to multiple substitutions (i.e., saturation effects), especially at third codon positions which are less constrained. This is also consistent with the bimodal distribution of variable sites in this dataset (Figure 4). Overall, these results show that data type and alignment information content have the potential to compromise phylogenomic inference from gene tree summarization methods that rely on the MSC model, even for datasets that contain relatively large amounts of data (Roch and Warnow 2015; Springer and Gatesy 2016).

We measured rates of gene tree estimation error using RF distances between gene trees and the inferred species tree, under the assumption that increases in RF distance were indicative of higher rates of phylogenetic error. Although gene histories are not expected to be congruent with the species tree in many cases (Tajima 1983; Pamilo and Nei 1988; Maddison 1997) , the assumption that RF distances are good indicators of error rates is valid because different datasets contained different means and variances in RF distances among gene trees. This was evident when the three UCE datasets were combined into a single analysis, which showed asymptotic convergence on relatively low mean and variance RF distances after reaching about 500 PIS (Figure 6). The asymptotic shape of this relationship implies that as PIS are added to an alignment, the resulting gene trees converge on a level of heterogeneity that is representative of or approaching real biological signal (i.e., ILS and/or introgression) rather than methodological or data-driven artifacts. Moreover, if all gene tree heterogeneity was due to biological processes such as ILS, one would expect the mean and variance in RF distances to be constant among datasets because these are derived from the same underlying samples. Although some level of gene tree heterogeneity is expected in all phylogenomic datasets, even if one assumes the impossible, where empirical gene tree estimation error is absent (Maddison 1997; Gatesy and Springer 2014; Edwards et al. 2016) , it is reasonable to assume that differences in the mean and variance in RF distances among datasets are good proxies for differences in gene tree estimation error.

Comparisons of estimation error between data types and the benefits of whole genome phylogenomics

We scrutinized orthologous data harvested from whole genome assemblies to identify the characteristics of phylogenomic datasets that might lead to lower bias. We found that short alignments with relatively few PIS had increased error (measured as difference between gene tree and species tree), a finding consistent with other studies (Meiklejohn et al. 2016; Burbrink et al. 2020) . Likewise, we found that long UCEs tend to have low rates of gene tree estimation error and are efficacious markers to use in phylogenomic studies. We also found that the gene trees from the protein-coding dataset derived from BUSCO harvesting and splicing of exons from the tinamou genomes were very noisy. Despite containing large numbers and proportions 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 of PIS (Figure 5) available across many genes, we found high variance in RF distances for these genes and thus conclude that such datasets might be more prone to error than datasets of long UCEs, which had lower gene tree estimation error and behaved more predictably (Figures 4 and 5). As a comparison, the UCE300Flank dataset also had high variance in RF distances (i.e., it contained a mix of low and high error trees), but this variance was well explained by information content. Thus, coding and UCE datasets have different underlying properties of molecular evolution, and these differences should be considered when selecting loci for study and in data analysis. Despite these differences, BUSCO and UCE300Flank datasets converged on the same phylogenetic topology for Clade A, perhaps related to their similar levels of gene tree heterogeneity (Wilcoxon rank sum tests accepted the null hypothesis that RF distances for both datasets did not differ).

Although we derived datasets from whole genome sequences, the UCE100Flank and UCE300Flank datasets are likely to most closely match datasets obtained from typical target capture approaches (Smith et al. 2014; Musher and Cracraft 2018; Tea et al. 2021) and are therefore indicative of rates of error in datasets derived from those protocols. For example, many studies utilize sequences from degraded DNA such as historical museum specimens, which may result in many relatively short UCE alignments. Still, even datasets sequenced from fresh tissues typically only capture about 100-300 bp of flanking region around the conserved UCE core. Thus, we offer multiple recommendations for future phylogenomic studies. First, if available, WGS data are highly preferred to sequence capture datasets because they allow for analysis of a much wider array of data types that can be compared as done herein (Jarvis et al. 2014; Zhang et al. 2019) . Such data also allow studies to achieve longer flanking regions around conserved UCE cores that significantly reduce bias during gene tree estimation (Figures 5 and 6), and enable analyses of many other processes of interest, such as evolution of gene families. Moreover, we did not include intron data in the current study, but these data would also be available from WGS and may behave more like UCE data rather than like the nearby exons themselves.

Second, if a very large genome size is cost prohibitive for WGS, then target capture datasets such as UCEs will likely contain high variance in RF distances (i.e., a mix of <good= and <bad= loci), but this variance is likely to be well explained by the number of PIS in each alignment (Figure 6). We therefore suggest that these datasets should be scrutinized and filtered, especially if a rapid radiation is involved (Doyle et al. 2015; S.A. Smith et al. 2018; Mclean et al. 2019; Smith et al. 2023) . As the number and proportion of PIS were strongly predictive of gene tree estimation error, we suggest that simply removing alignments with too few PIS may be useful in many cases, as has been suggested elsewhere (Harris et al. 2014; Hosner et al. 2016; Blom et al. 2017; Leite et al. 2021) . Given these findings, common phylogenomic approaches that sequence loci with intermediate or low levels of information content such as sequence capture of UCEs may be insufficient to resolve the relationships of taxa that evolved rapidly, such as adaptive radiations, where rates of ILS and short internodal distances caused by rapid diversification are expected to be extreme (Gatesy and Springer 2014; Mclean et al. 2019) . In these situations, the number of variable sites may also be too few to resolve gene trees involving short branches.

Finally, although our BUSCO dataset was noisy and included a range of high- and lowerror gene trees, we do not suggest that datasets of coding genes are inherently poor quality. 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461

Indeed, protein-coding datasets such as BUSCOs, transcriptomes, and others are likely to remain useful for a range of phylogenomic problems. For example, given their conservation of function, datasets of protein-coding genes will be important for reconstructing phylogenetic relationships at deeper timescales using either nucleotide or amino acid sequences (Dunn et al. 2008; Allen et al. 2017; Zhang et al. 2019; Boyd et al. 2022) . Still, given that our BUSCO dataset was noisy compared with our UCE datasets, we suggest exercising caution when applying protein-coding genes to difficult phylogenomic problems (Reddy et al. 2017). Careful fitting of substitution models, data filtering, and perhaps even manual inspection of alignments may reduce bias in these and other datasets of protein-coding sequence (Anisimova and Kosiol 2009; Springer and Gatesy 2016) .

However, our results do not suggest that BUSCOs, or protein-coding genes in general, should be filtered by information content or RF distances. Information content was a weak and inconsistent predictor of gene tree estimation error for BUSCOs. Alignments with a fewer number of PIS or a higher proportion of PIS were both associated with increased gene tree error. Likewise, some amount of gene tree heterogeneity is expected due to the coalescent process. Thus, assuming that all high RF gene trees are erroneous and removing them from a dataset could mislead MSC analyses. Thus, more research is needed to understand how best to filter BUSCO datasets, and perhaps protein-coding datasets in general, for phylogenomics using gene tree summarization methods.

We thank Alvaro Hernandez and Chris Wright at the University of Illinois Roy J. Carver Biotechnology Center for assistance with Illumina sequencing. We thank Kimberly K. O. Walden for assistance in depositing Illumina reads to NCBI SRA. We also thank Jorge Doña for providing R code to assist with breakpoint analysis. Dan Lane, Steve Cardiff, and Christopher Witt helped confirm voucher ID9s for multiple samples. We thank B. T. Smith, J. Cracraft, P. Sweet, T. Trombone, S. Katanova (AMNH), J. Bates, S. Hackett, and B. Marks (FMNH), F. Sheldon and D. Dittman (LSUMNS), H.F. James and B. Schmidt (USNM), and N. Rice (ANSP) for loaning material that was crucial for this work. Finally, we thank the members of ANSP Ornithology Department (A. Del Grosso, E. Griffith, K. Kuabara, and J. Merwin) for comments on early versions of the manuscript. This study was funded by NSF grant DEB-1855812 to JDW and KPJ. TAC was partially supported by NSF grant DEB-2203228.

All raw reads are available on the NCBI Sequence Read Archive (project and sample accession numbers can be found in Table S1). All data and code needed to replicate the analyses found herein can be found at LJM9s github page: https://github.com/lukemusher/Tinamou-phylogenomics.git

Abadi S, Azouri D, Pupko T, Mayrose I. 2019. Model selection may not be a mandatory step for phylogeny reconstruction. Nat. Commun. 10:934.

Alaei Kakhki N, Schweizer M, Lutgen D, Bowie RCK, Shirihai H, Suh A, Schielzeth H, Burri R. 2023. A Phylogenomic Assessment of Processes Underpinning Convergent Evolution in Open-Habitat Chats. Mol. Biol. Evol. [Internet] 40. Available from: 680 Allen JM, Boyd B, Nguyen N-P, Vachaspati P, Warnow T, Huang DI, Grady PGS, Bell KC, Cronk QCB, Mugisha L, et al. 2017. Phylogenomics from Whole Genome Sequences Using aTRAM. Syst. Biol. 66:786–798.

Allman ES, Baños H, Mitchell JD, Rhodes JA. 2023. MSCquartets: analyzing gene tree quartets under the multi-species coalescent. R package version 1.3.1. Available from: https://CRAN.R-project.org/package=MSCquartets Almeida FC, Porzecanski AL, Cracraft JL, Bertelli S. 2022. The evolution of tinamous (Palaeognathae: Tinamidae) in light of molecular and combined analyses. Zool. J. Linn.

Soc. 195:106–124.

Altimiras J, Lindgren I, Giraldo-Deck LM, Matthei A, Garitano-Zavala Á. 2017. Aerobic performance in tinamous is limited by their small heart. A novel hypothesis in the evolution of avian flight. Sci. Rep. 7:15964.

Anisimova M, Kosiol C. 2009. Investigating protein-coding sequence evolution with probabilistic codon substitution models. Mol. Biol. Evol. 26:255–271.

Bertelli S. 2017. Advances on tinamou phylogeny: an assembled cladistic study of the volant palaeognathous birds. Cladistics 33:351–374.

Bertelli S, Giannini NP, Goloboff PA. 2002. A phylogeny of the tinamous (aves: palaeognathiformes) based on integumentary characters. Syst. Biol. 51:959–979. Bertelli S, Porzecanski AL. 2004. Tinamou (tinamidae) systematics: A preliminary combined analysis of morphology and molecules. Ornitol. Neotrop. 15 (Suppl):293–299. Blom MPK, Bragg JG, Potter S, Moritz C. 2017. Accounting for Uncertainty in Gene Tree Estimation: Summary-Coalescent Species Tree Inference in a Challenging Radiation of Australian Lizards. Syst. Biol. 66:352–366.

Boyd BM, Nguyen N-P, Allen JM, Waterhouse RM, Vo KB, Sweet AD, Clayton DH, Bush SE, Shapiro MD, Johnson KP. 2022. Long-distance dispersal of pigeons and doves generated new ecological opportunities for host-switching and adaptive radiation by their parasites.

Proc. Biol. Sci. 289:20220042.

Brinkmann H, van der Giezen M, Zhou Y, Poncelin de Raucourt G, Philippe H. 2005. An empirical assessment of long-branch attraction artefacts in deep eukaryotic phylogenomics.

Syst. Biol. 54:743–757. 719

Accipiter hawks in the Caribbean islands. The Auk 138:1–23. 757 758 759 760 761 762

Gueuning M, Frey JE, Praz C. 2020. Ultraconserved yet informative for species delimitation: Ultraconserved elements resolve long-standing systematic enigma in Central European bees. Mol. Ecol. 29:4203–4220.

Harris RB, Carling MD, Lovette IJ. 2014. The influence of sampling design on species tree inference: a new relationship for the New World chickadees (Aves: Poecile). Evolution 68:501–513. 796 797 798 799 800 801 835 836 837 838

Nature 526:569–573.

Ranwez V, Douzery EJP, Cambon C, Chantret N, Delsuc F. 2018. MACSE v2: Toolkit for the Alignment of Coding Sequences Accounting for Frameshifts and Stop Codons. Mol. Biol.

Evol. 35:2582–2584. Reddy S, Kimball RT, Pandey A, Hosner PA, Braun MJ, Hackett SJ, Han K-L, Harshman J, Huddleston CJ, Kingston S, et al. 2017. Why Do Phylogenomic Data Sets Yield Conflicting Trees? Data Type Influences the Avian Tree of Life more than Taxon Sampling. Syst. Biol. 66:857–879.

Rhodes JA, Baños H, Mitchell JD, Allman ES. 2021. MSCquartets 1.0: quartet methods for species trees and networks under the multispecies coalescent model in R. Bioinformatics 37:1766–1768.

Robinson DF, Foulds LR. 1981. Comparison of phylogenetic trees. Math. Biosci. 53:131–147. Roch S, Warnow T. 2015. On the Robustness to Gene Tree Estimation Error (or lack thereof) of

Coalescent-Based Species Tree Methods. Syst. Biol. 64:663–676.

Sackton TB, Grayson P, Cloutier A, Hu Z, Liu JS, Wheeler NE, Gardner PP, Clarke JA, Baker AJ, Clamp M, et al. 2019. Convergent regulatory evolution and loss of flight in paleognathous birds. Science 364:74–78.

Schliep KP. 2011. phangorn: phylogenetic analysis in R. Bioinformatics 27:592–593. Schultz DT, Haddock SHD, Bredeson JV, Green RE, Simakov O, Rokhsar DS. 2023. Ancient gene linkages support ctenophores as sister to other animals. Nature 618:110–117. Simão FA, Waterhouse RM, Ioannidis P, Kriventseva EV, Zdobnov EM. 2015. BUSCO: assessing genome assembly and annotation completeness with single-copy orthologs.

Bioinformatics 31:3210–3212.

Simion P, Philippe H, Baurain D, Jager M, Richter DJ, Di Franco A, Roure B, Satoh N, Quéinnec É, Ereskovsky A, et al. 2017. A Large and Consistent Phylogenomic Dataset Supports Sponges as the Sister Group to All Other Animals. Curr. Biol. 27:958–967. Smith BT, Harvey MG, Faircloth BC, Glenn TC, Brumfield RT. 2014. Target capture and massively parallel sequencing of ultraconserved elements for comparative studies at shallow evolutionary time scales. Syst. Biol. 63:83–95.

Smith BT, Mauck WM, Benz B, Andersen MJ. 2018. Uneven missing data skews phylogenomic relationships within the lories and lorikeets [Internet].

Smith BT, Merwin J, Provost KL, Thom G, Brumfield RT, Ferreira M, Mauck WM, Moyle RG, Wright TF, Joseph L. 2023. Phylogenomic Analysis of the Parrots of the World Distinguishes Artifactual from Biological Sources of Gene Tree Discordance. Syst. Biol. 72:228–241.

Smith SA, Brown JW, Walker JF. 2018. So many genes, so little time: A practical approach to 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911

divergence-time estimation in the genomic era. PLoS One 13:e0197433.

Springer MS, Gatesy J. 2016. The gene tree delusion. Mol. Phylogenet. Evol. 94:1–33. Stamatakis A. 2006. RAxML-VI-HPC: maximum likelihood-based phylogenetic analyses with thousands of taxa and mixed models. Bioinformatics 22:2688–2690.

Stamatakis A. 2014. RAxML version 8: a tool for phylogenetic analysis and post-analysis of large phylogenies. Bioinformatics 30:1312–1313.

Tajima F. 1983. Evolutionary relationship of DNA sequences in finite populations. Genetics 105:437–460. 912 913 914 915 916

Zhao M, Kurtis SM, White ND, Moncrieff AE, Leite RN, Brumfield RT, Braun EL, Kimball RT. (Aves: Pipridae). Syst. Biol. 72:161–178. 918

Dataset BUSCOs UCE100Flank UCE300Flank

Mean #PIS

StDev #PIS

Mean %PIS

StDev %PIS

Mean RF

StDev RF Robinson-Foulds Distances (RF) for each dataset analyzed for the study.

Figure 1: Concatenated Phylogeny of tinamous based on UCEs with 1,000 bp of flanking sequence. All nodes have bootstrap value of 100% except where noted. Note relatively short internodal branch lengths in Clade A. Tinamou illustrations by TAC. with 1,000 bp of flanking sequence. All nodes have posterior probability of 1. Branches labeled B1–B5 denote rapidly diverging branches evaluated using quartet analysis (see Figure 7). Tinamou illustrations by TAC. 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 regression models exploring the relationship between information content and gene tree heterogeneity. The negative correlation in most instances suggests that alignments with less information content result in increased gene tree estimation error. 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989